As of November 4, 2026, Level lists 60 live roles with post-training in the title. Of these, 58 sit at Builds AI. A fresh debate about editing model weights has put this stage of model building in the news. The roles are open anyway. Of those that list pay, the median lower bound is $250,000+.
What happened
Yesterday we covered an MIT CSAIL post on weight-space programming. The post says a language model can read neural network weights and edit them to fix a small model.
The pushback came fast. Simon Willison wrote that a Qwen model on his laptop can interpret model weights too. Matthew Honnibal argued the original framing was misleading. Sean Goedecke posted that LLMs are not superhuman at reading and writing weights.
Level did not reproduce any of these results. We link them so you can judge the claims yourself. What matters for a job seeker is narrower. Editing weights after training is one more tool in the stage that teams call post-training.
What are post-training jobs?
Post-training is the work done after a model's first large training run. It includes fine-tuning, reinforcement learning from human feedback, preference tuning and evaluation. The goal is to make a raw model useful, safe and steady.
A weight edit is a very small version of the same idea. Instead of retraining, you change the model after the fact. Teams that do this work are the people hiring under the titles below.
Day to day, that can mean designing reward signals, building test sets that catch regressions, or tuning how a model follows instructions. Listings differ a lot, so read the duties and not only the title.
Reinforcement learning appears in 11 titles. Evaluation appears in only 2, such as Thinking Machines Lab's Research, Post-Training Evals. The rest use broader titles like Member of Technical Staff, Post-Training.
Who is hiring for post-training jobs?
These roles come from 32 companies. NVIDIA and Waymo lead with 4 each. Anthropic, Cohere, Mercor, Mirendil, OpenAI and SonarSource post 3 each.
Level's AI levels show how central the work is. Most sit at Builds AI, where the job is building AI systems. Only 2 sit at Works on AI, where the job is working on AI systems built by others.
Remote work is limited here. Exactly 20 roles are flagged as remote, and remote AI jobs lists the wider set. Cohere's Member of Technical Staff, Post-Training is one of them.
What do post-training jobs pay?
Pay data is partial, so read it as a guide. Only 33 listings publish a yearly salary in US dollars. Their median lower bound is $250,000+.
The median upper bound is $356,000+. The highest posted upper bound is $850,000+. For a wider comparison, see our guide to research scientist salaries.
Can you start without a senior title?
Senior and lead roles make up 31 listings. Another 10 are labeled mid level.
Only 3 roles are labeled intern. If you are earlier in your career, look at AI research jobs and build evaluation experience. Evaluation experience is one place to start, though we did not measure hiring paths.
A practical step is to read three or four listings side by side and note which duties repeat. Repeated duties give you a checklist for your own portfolio.
Tool names add some detail. PyTorch shows up in 34 listings. Python appears in 25. JAX appears in 13. Tools show what listings mention, not what they require.
Open roles right now
This list uses Level search for post-training. It shows the closest matches, ranked by AI level. Check each listing for its own location and pay details before you apply.
